WAPR

Worldwide AI Proliferation Registry
operator register built 2026-11-01
build …latest run build f7813e9identity detector v4condition rules WCI-1
Operator record · derived from OpenStreetMap attribution

Microsoft

134 mapped facilities in the WAPR frame

feed checked 2026-11-01 09:17 UTC operator register built 2026-11-01

WAPR is a public register of where AI computing physically sits. This page is one company and the mapped buildings that public records attribute to it. Who runs each data centre, according to volunteer mapmakers. Where nobody has recorded an operator, this page says so instead of guessing — that silence is itself one of the findings.

mapped facilities
134
by country
  1. United States of America70
  2. Ireland12
  3. Netherlands12
  4. Sweden8
  5. Australia6
  6. Indonesia4
  7. United Kingdom3
  8. Chile2
  9. China2
  10. Denmark2
  11. New Zealand2
  12. Spain2
  13. Austria1
  14. Costa Rica1
  15. Dominican Rep.1
  16. France1
  17. Italy1
  18. Lebanon1
  19. Puerto Rico1
  20. Singapore1
  21. Trinidad and Tobago1
Counts of the buildings OpenStreetMap volunteers have mapped and attributed to Microsoft — not of everything Microsoft operates. How thoroughly volunteers map varies enormously from country to country, so these are not shares of the company's real footprint.
Wikidata
Q2283
How this is known

Map extracts

from country- or continent-wide extracts
  1. US42
  2. EUROPE41
  3. ASIA7
  4. AUSTRALIA-OCEANIA5
  5. CENTRAL-AMERICA3
  6. SOUTH-AMERICA2
from country extracts
  1. AT1
  2. SG1
from state or province extracts
  1. VA8
  2. TX6
  3. AZ5
  4. IL4
  5. GA3
  6. VIC3
  7. CA1
  8. WA1
  9. PR1
These are the map extracts each building was pulled from, not a geographic hierarchy — a building counted under a state is not also counted under its country. The levels do not nest, and this is not a breakdown by country.

Basin water-stress exposure

checked 2026-11-01 09:17 UTC data 1979–2019 baseline the basin layer’s own date is not stated in the payload; the figures beside it are from the run above

52 of 134 assessed sites (39%) sit in High or Extremely High baseline water-stress basins.

Extremely High (>80%)
21
High (40–80%)
31
Medium–High (20–40%)
4
Low–Medium (10–20%)
28
Low (<10%)
50

The fill steps from the strongest ink for Extremely High to the faintest for Low; each bar is that class’s share of all this operator’s mapped facilities, with the count beside it. The percentages bound WRI Aqueduct’s baseline water-stress classes: how much of a basin’s available water is already claimed (“Basin water stress” in the terms below).

This counts how many of a company's mapped buildings sit in parts of the world where fresh water is already scarce. It does NOT measure how much water they use — almost no operator publishes that. Each building counts once, big or small, because reliable sizes are not public. Built entirely from public maps and public hydrology; nothing a company says about itself is used.

Count-weighted exposure (per-site capacity is not publishable); OSM attribution covers only attributed sites; exposure is not intake. Chain: OSM mapping x WRI Aqueduct 4.0 basins (CC BY 4.0) — no corporate disclosure enters these figures.

Mapped facilities (134 listed)

Show the other 110

View on the Explorer globe

Attribution comes from what OpenStreetMap contributors recorded, resolved through a deterministic alias table. Facilities whose operator no one recorded are absent here — absence of attribution is not absence of facilities. Water figures are basin exposure, never measured intake.